[feature]Added support for PaddlePaddle model artifact (#1523)

.github/workflows/ci.yml

@@ -144,6 +144,25 @@ jobs:
       - name: Upload test coverage to Codecov
         uses: codecov/codecov-action@v1.0.12
 
+  paddle_integration_tests:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+        with: 
+          fetch-depth: 0
+      - name: Setup python
+        uses: actions/setup-python@v2
+        with:
+          python-version: 3.8
+      - name: Install test dependencies
+        run: ./ci/install_test_deps.sh
+      - name: Install paddle
+        run: python -m pip install paddlepaddle
+      - name: Run tests
+        run: ./ci/test_project.sh tests/integration/projects/paddle
+      - name: Upload test coverage to Codecov
+        uses: codecov/codecov-action@v1.0.12
+
   api_server_integration_tests:
     name: API Server Integration Tests (${{ matrix.os }})
     runs-on: ${{ matrix.os }}

README.md

@@ -94,7 +94,7 @@ Standarlize model serving and deployment workflow for teams:
 * Transformers - [Docs](https://docs.bentoml.org/en/latest/frameworks.html#transformers)
 * Gluon - [Docs](https://docs.bentoml.org/en/latest/frameworks.html#gluon)
 * Detectron - [Docs](https://docs.bentoml.org/en/latest/frameworks.html#detectron)
-
+* Paddle - [Docs](https://docs.bentoml.org/en/latest/frameworks.html#paddle)
 
 
 ### Deployment Options

bentoml/artifact/__init__.py

@@ -41,6 +41,7 @@
 from bentoml.frameworks.spacy import SpacyModelArtifact
 from bentoml.frameworks.tensorflow import TensorflowSavedModelArtifact
 from bentoml.frameworks.xgboost import XgboostModelArtifact
+from bentoml.frameworks.paddle import PaddlePaddleModelArtifact  # noqa: E402
 
 __all__ = [
     "ArtifactCollection",
@@ -64,4 +65,5 @@
     "TextFileArtifact",
     "XgboostModelArtifact",
     "PytorchLightningModelArtifact",
+    "PaddlePaddleModelArtifact",
 ]

bentoml/frameworks/paddle.py

@@ -0,0 +1,108 @@
+import os
+
+from bentoml.exceptions import MissingDependencyException
+from bentoml.service.artifacts import BentoServiceArtifact
+from bentoml.service.env import BentoServiceEnv
+import tempfile
+
+
+try:
+    import paddle
+    import paddle.inference as paddle_infer
+except ImportError:
+    paddle = None
+
+
+class PaddlePaddleModelArtifact(BentoServiceArtifact):
+    """Abstraction for saving/loading PaddlePaddle models
+
+    Args:
+        name (string): name of the artifact
+
+    Raises:
+        MissingDependencyException: paddle package is required for
+                                    PaddlePaddleModelArtifact
+
+    Example usage:
+
+    >>> import pandas as pd
+    >>>
+    >>> from bentoml import env, artifacts, api, BentoService
+    >>> from bentoml.adapters import DataframeInput
+    >>> from bentoml.frameworks.paddle import PaddlePaddleModelArtifact
+    >>>
+    >>> @env(infer_pip_packages=True)
+    >>> @artifacts([PaddlePaddleModelArtifact('model')])
+    >>> class PaddleService(BentoService):
+    >>>    @api(input=DataframeInput(), batch=True)
+    >>>    def predict(self, df: pd.DataFrame):
+    >>>        input_data = df.to_numpy().astype(np.float32)
+    >>>        predictor = self.artifacts.model
+    >>>
+    >>>        input_names = predictor.get_input_names()
+    >>>        input_handle = predictor.get_input_handle(input_names[0])
+    >>>        input_handle.reshape(input_data.shape)
+    >>>        input_handle.copy_from_cpu(input_data)
+    >>>
+    >>>        predictor.run()
+    >>>
+    >>>        output_names = predictor.get_output_names()
+    >>>        output_handle = predictor.get_output_handle(output_names[0])
+    >>>        output_data = output_handle.copy_to_cpu()
+    >>>        return output_data
+    >>>
+    >>> service = PaddleService()
+    >>>
+    >>> service.pack('model', model_to_save)
+    """
+
+    def __init__(self, name: str):
+        super(PaddlePaddleModelArtifact, self).__init__(name)
+        self._model = None
+        self._predictor = None
+        self._model_path = None
+
+        if paddle is None:
+            raise MissingDependencyException(
+                "paddlepaddle package is required to use PaddlePaddleModelArtifact"
+            )
+
+    def pack(self, model):  # pylint:disable=arguments-differ
+        self._model = model
+        return self
+
+    def load(self, path):
+        model = paddle.jit.load(self._file_path(path))
+        model = paddle.jit.to_static(model, input_spec=model._input_spec())
+        return self.pack(model)
+
+    def _file_path(self, base_path):
+        return os.path.join(base_path, self.name)
+
+    def save(self, dst):
+        self._save(dst)
+
+    def _save(self, dst):
+        # Override the model path if temp dir was set
+        self._model_path = self._file_path(dst)
+        paddle.jit.save(self._model, self._model_path)
+
+    def get(self):
+        # Create predictor, if one doesn't exist, when inference is run
+        if not self._predictor:
+            # If model isn't saved, save model to a temp dir
+            # because predictor init requires the path to a saved model
+            if self._model_path is None:
+                self._model_path = tempfile.TemporaryDirectory().name
+                self._save(self._model_path)
+
+            config = paddle_infer.Config(
+                self._model_path + ".pdmodel", self._model_path + ".pdiparams"
+            )
+            config.enable_memory_optim()
+            predictor = paddle_infer.create_predictor(config)
+            self._predictor = predictor
+        return self._predictor
+
+    def set_dependencies(self, env: BentoServiceEnv):
+        env.add_pip_packages(['paddlepaddle'])

docs/source/frameworks.rst

@@ -204,6 +204,17 @@ Detectron
 .. autoclass:: bentoml.frameworks.detectron.DetectronModelArtifact
 
 
+=========
+Paddle
+=========
+
+Example Projects:
+
+* Boston Housing Prediction - `Google Colab <https://colab.research.google.com/github/bentoml/gallery/blob/master/paddlepaddle/LinearRegression/LinearRegression.ipynb>`__ / `Notebook Source <https://github.com/bentoml/gallery/blob/master/paddlepaddle/LinearRegression/LinearRegression.ipynb>`__
+
+.. autoclass:: bentoml.frameworks.paddle.PaddleModelArtifact
+
+
 .. spelling::
 
     MLModel

tests/integration/projects/paddle/model/model.py

@@ -0,0 +1,58 @@
+import pathlib
+import sys
+
+import paddle
+from paddle import nn
+from paddle.static import InputSpec
+from bentoml.frameworks.paddle import PaddlePaddleModelArtifact
+
+
+BATCH_SIZE = 8
+BATCH_NUM = 4
+EPOCH_NUM = 5
+
+IN_FEATURES = 13
+OUT_FEATURES = 1
+
+
+class Model(nn.Layer):
+    def __init__(self):
+        super(Model, self).__init__()
+        self.fc = nn.Linear(IN_FEATURES, OUT_FEATURES)
+
+    @paddle.jit.to_static(input_spec=[InputSpec(shape=[IN_FEATURES], dtype='float32')])
+    def forward(self, x):
+        return self.fc(x)
+
+
+def train(model, loader, loss_fn, opt):
+    model.train()
+    for epoch_id in range(EPOCH_NUM):
+        for batch_id, (feature, label) in enumerate(loader()):
+            out = model(feature)
+            loss = loss_fn(out, label)
+            loss.backward()
+            opt.step()
+            opt.clear_grad()
+
+
+def pack_models(path):
+    model = Model()
+    loss = nn.MSELoss()
+    adam = paddle.optimizer.Adam(parameters=model.parameters())
+
+    train_data = paddle.text.datasets.UCIHousing(mode="train")
+
+    loader = paddle.io.DataLoader(
+        train_data, batch_size=BATCH_SIZE, shuffle=True, drop_last=True, num_workers=2
+    )
+
+    train(model, loader, loss, adam)
+
+    PaddlePaddleModelArtifact("model").pack(model).save(path)
+
+
+if __name__ == "__main__":
+    artifacts_path = sys.argv[1]
+    pathlib.Path(artifacts_path).mkdir(parents=True, exist_ok=True)
+    pack_models(artifacts_path)

tests/integration/projects/paddle/service.py

@@ -0,0 +1,43 @@
+import pathlib
+import sys
+
+import numpy as np
+
+import bentoml
+from bentoml.adapters import DataframeInput
+from bentoml.frameworks.paddle import PaddlePaddleModelArtifact
+
+
+@bentoml.env(infer_pip_packages=True)
+@bentoml.artifacts([PaddlePaddleModelArtifact('model')])
+class PaddleService(bentoml.BentoService):
+    @bentoml.api(input=DataframeInput(), batch=True)
+    def predict(self, df):
+        input_data = df.to_numpy().astype(np.float32)
+
+        predictor = self.artifacts.model
+        input_names = predictor.get_input_names()
+        input_handle = predictor.get_input_handle(input_names[0])
+
+        input_handle.reshape(input_data.shape)
+        input_handle.copy_from_cpu(input_data)
+
+        predictor.run()
+
+        output_names = predictor.get_output_names()
+        output_handle = predictor.get_output_handle(output_names[0])
+        output_data = output_handle.copy_to_cpu()
+        return output_data
+
+
+if __name__ == "__main__":
+    artifacts_path = sys.argv[1]
+    bento_dist_path = sys.argv[2]
+    service = PaddleService()
+
+    from model.model import Model  # noqa # pylint: disable=unused-import
+
+    service.artifacts.load_all(artifacts_path)
+
+    pathlib.Path(bento_dist_path).mkdir(parents=True, exist_ok=True)
+    service.save_to_dir(bento_dist_path)

tests/integration/projects/paddle/tests/test_service.py

@@ -0,0 +1,28 @@
+import pandas as pd
+import numpy as np
+
+test_df = pd.DataFrame(
+    [
+        [
+            -0.0405441,
+            0.06636364,
+            -0.32356227,
+            -0.06916996,
+            -0.03435197,
+            0.05563625,
+            -0.03475696,
+            0.02682186,
+            -0.37171335,
+            -0.21419304,
+            -0.33569506,
+            0.10143217,
+            -0.21172912,
+        ]
+    ]
+)
+
+
+def test_paddle_artifact_pack(service):
+    pred = service.predict(test_df)
+    assert isinstance(pred, np.ndarray), 'Run inference'
+    assert pred.shape == (1, 1)